Reducing network related complaints has always been one of the important work of telecom operators. At present

most of the early warning schemes for network complaint users are based on the experience of network optimization engineers

with low accuracy and efficiency. In this paper

through a comprehensive and in-depth analysis of the historical network complaint user data

a complaint user feature model is build based on XGboost algorithm

which can predict network complaint users. This method has a high prediction accuracy

and can locate the reasons for poor quality of users after docking with other network optimization systems

so that the network department can optimize the network in advance and improve user satisfaction.